OpenAI Says It’s "Over" If It Can’t Steal All Your Copyrighted Work
-
can you point to the trial they won? I only know about a case that was dismissed.
because what we've seen from ai so far is hardly transformative.
Sorry, I was talking about HiQ labs v. Linkedin. But there is Google v. Perfect 10 and Google v. Authors Guild that show how scrapping public data is perfectly fine and include Google.
An image generator is trained on a billion images and is able to spit out completely new images on whatever you ask it. Calling it anything but transformative is silly, especially when such things as collage are considered transformative.
-
Sorry, I was talking about HiQ labs v. Linkedin. But there is Google v. Perfect 10 and Google v. Authors Guild that show how scrapping public data is perfectly fine and include Google.
An image generator is trained on a billion images and is able to spit out completely new images on whatever you ask it. Calling it anything but transformative is silly, especially when such things as collage are considered transformative.
eh, "completely new" is a huge stretch there. splicing two or ten movies together doesn't give you an automatic pass.
-
This is a tough one
Open-ai is full of shit and should die but then again, so should copyright law as it currently is
That's fair, but OpenAI isn't fighting to reform copyright law for everyone. OpenAI wants you to be subject to the same restrictions you currently face, and them to be exempt. This isn't really a "lesser of two evils" situation.
-
This post did not contain any content.
Okay, I can work with this. Hey Altman you can train on anything that's public domain, now go take those fuck ton of billions and fight the copyright laws to make public domain make sense again.
-
This post did not contain any content.
Too bad, so sad
-
Also true. It’s scraping.
In the words of Cory Doctorow:
Web-scraping is good, actually.
Scraping against the wishes of the scraped is good, actually.
Scraping when the scrapee suffers as a result of your scraping is good, actually.
Scraping to train machine-learning models is good, actually.
Scraping to violate the public’s privacy is bad, actually.
Scraping to alienate creative workers’ labor is bad, actually.
We absolutely can have the benefits of scraping without letting AI companies destroy our jobs and our privacy. We just have to stop letting them define the debate.
Molly White also wrote about this in the context of open access on the web and people being concerned about how their works are being used.
“Wait, not like that”: Free and open access in the age of generative AI
The same thing happened again with the explosion of generative AI companies training models on CC-licensed works, and some were disappointed to see the group take the stance that, not only do CC licenses not prohibit AI training wholesale, AI training should be considered non-infringing by default from a copyright perspective.
-
This has been the legal basis of all AI training sets since they began collecting datasets. The US copyright office heard these arguments in 2023: https://www.copyright.gov/ai/listening-sessions.html
MR. LEVEY: Hi there. I'm Curt Levey, President of the Committee for Justice. We're a nonprofit that focuses on a variety of legal and policy issues, including intellectual property, AI, tech policy. There certainly are a number of very interesting questions about AI and copyright. I'd like to focus on one of them, which is the intersection of AI and copyright infringement, which some of the other panelists have already alluded to.
That issue is at the forefront given recent high-profile lawsuits claiming that generative AI, such as DALL-E 2 or Stable Diffusion, are infringing by training their AI models on a set of copyrighted images, such as those owned by Getty Images, one of the plaintiffs in these suits. And I must admit there's some tension in what I think about the issue at the heart of these lawsuits. I and the Committee for Justice favor strong protection for creatives because that's the best way to encourage creativity and innovation.
But, at the same time, I was an AI scientist long ago in the 1990s before I was an attorney, and I have a lot of experience in how AI, that is, the neural networks at the heart of AI, learn from very large numbers of examples, and at a deep level, it's analogous to how human creators learn from a lifetime of examples. And we don't call that infringement when a human does it, so it's hard for me to conclude that it's infringement when done by AI.
Now some might say, why should we analogize to humans? And I would say, for one, we should be intellectually consistent about how we analyze copyright. And number two, I think it's better to borrow from precedents we know that assumed human authorship than to invent the wheel over again for AI. And, look, neither human nor machine learning depends on retaining specific examples that they learn from.
So the lawsuits that I'm alluding to argue that infringement springs from temporary copies made during learning. And I think my number one takeaway would be, like it or not, a distinction between man and machine based on temporary storage will ultimately fail maybe not now but in the near future. Not only are there relatively weak legal arguments in terms of temporary copies, the precedent on that, more importantly, temporary storage of training examples is the easiest way to train an AI model, but it's not fundamentally required and it's not fundamentally different from what humans do, and I'll get into that more later if time permits.
The "temporary copy" idea is pretty central for visual models like Midjourney or DALL-E, whose training sets are full of copyrighted works lol. There is a legal basis for temporary copies too:
The "Ephemeral Copy" Exception (17 U.S.C. § 112 & § 117)
U.S. copyright law recognizes temporary, incidental, and transitory copies as necessary for technological processes. Section 117 allows temporary copies for software operation. Section 112 permits temporary copies for broadcasting and streaming.BTW, if anyone was interested - many visual models use the same training set, collected by a German non-profit: https://laion.ai/
It's "technically not copyright infringement" because the set is just a link to an image, paired with a text description of each image. Because they're just pointing to the image, they don't really have to respect any copyright.
-
What I’m hearing between the lines here is the origin of a legal “argument.”
If a person’s mind is allowed to read copyrighted works, remember them, be inspired by them, and describe them to others, then surely a different type of “person’s” different type of “mind” must be allowed to do the same thing!
After all, corporations are people, right? Especially any worth trillions of dollars! They are more worthy as people than meatbags worth mere billions!
I don't think it's actually such a bad argument because to reject it you basically have to say that style should fall under copyright protections, at least conditionally, which is absurd and has obvious dystopian implications. This isn't what copyright was meant for. People want AI banned or inhibited for separate reasons and hope the copyright argument is a path to that, but even if successful wouldn't actually change much except to make the other large corporations that own most copyright part owners of AI systems. That's not actually a better circumstance.
-
Yeah, he has the ability to articulate what I was already thinking about LLMs and bring in hard data to back up his thesis that it’s all bullshit. Dangerous and expensive bullshit, but bullshit nonetheless.
It’s really sad that his willingness to say the tech industry is full of shit is such an unusual attribute in the tech journalism world.
It’s really sad that his willingness to say the tech industry is full of shit is such an unusual attribute in the tech journalism world.
What is interesting is if he didn't pretty regularly say "why the fuck AM I the guy who is sounding the alarm here????!?!?!" I would be much more skeptical of his points. He isn't someone that is directly aligned with the industry, at least not in an "authoritative expert capable of doing a thorough takedown of a bubble/hype mirage". I mean I can tell the guy likes the attention, but he seems utterly genuine in the "wtf, well ok I will do it... but like seriously I AM the guy who is sounding the alarm here? This isn't honestly my direct area of expertise?"
-
I feel like it would be ok if AI generated images/text would be clearly marked(but i dont think its possible in the case of text)
Who would support something made stealing the hard work of other people if they could tell instantly
Stealing means the initial item is no longer there
-
Has been since 1991
Took a brief break for MENA to be the targeted one though
-
Okay, I can work with this. Hey Altman you can train on anything that's public domain, now go take those fuck ton of billions and fight the copyright laws to make public domain make sense again.
This is the correct answer. Never forget that US copyright law originally allowed for a 14 year (renewable for 14 more years) term. Now copyright holders are able to:
- reach consumers more quickly and easily using the internet
- market on more fronts (merch didn't exist in 1710)
- form other business types to better hold/manage IP
So much in the modern world exists to enable copyright holders, but terms are longer than ever. It's insane.
-
I don't think it's actually such a bad argument because to reject it you basically have to say that style should fall under copyright protections, at least conditionally, which is absurd and has obvious dystopian implications. This isn't what copyright was meant for. People want AI banned or inhibited for separate reasons and hope the copyright argument is a path to that, but even if successful wouldn't actually change much except to make the other large corporations that own most copyright part owners of AI systems. That's not actually a better circumstance.
-
This has been the legal basis of all AI training sets since they began collecting datasets. The US copyright office heard these arguments in 2023: https://www.copyright.gov/ai/listening-sessions.html
MR. LEVEY: Hi there. I'm Curt Levey, President of the Committee for Justice. We're a nonprofit that focuses on a variety of legal and policy issues, including intellectual property, AI, tech policy. There certainly are a number of very interesting questions about AI and copyright. I'd like to focus on one of them, which is the intersection of AI and copyright infringement, which some of the other panelists have already alluded to.
That issue is at the forefront given recent high-profile lawsuits claiming that generative AI, such as DALL-E 2 or Stable Diffusion, are infringing by training their AI models on a set of copyrighted images, such as those owned by Getty Images, one of the plaintiffs in these suits. And I must admit there's some tension in what I think about the issue at the heart of these lawsuits. I and the Committee for Justice favor strong protection for creatives because that's the best way to encourage creativity and innovation.
But, at the same time, I was an AI scientist long ago in the 1990s before I was an attorney, and I have a lot of experience in how AI, that is, the neural networks at the heart of AI, learn from very large numbers of examples, and at a deep level, it's analogous to how human creators learn from a lifetime of examples. And we don't call that infringement when a human does it, so it's hard for me to conclude that it's infringement when done by AI.
Now some might say, why should we analogize to humans? And I would say, for one, we should be intellectually consistent about how we analyze copyright. And number two, I think it's better to borrow from precedents we know that assumed human authorship than to invent the wheel over again for AI. And, look, neither human nor machine learning depends on retaining specific examples that they learn from.
So the lawsuits that I'm alluding to argue that infringement springs from temporary copies made during learning. And I think my number one takeaway would be, like it or not, a distinction between man and machine based on temporary storage will ultimately fail maybe not now but in the near future. Not only are there relatively weak legal arguments in terms of temporary copies, the precedent on that, more importantly, temporary storage of training examples is the easiest way to train an AI model, but it's not fundamentally required and it's not fundamentally different from what humans do, and I'll get into that more later if time permits.
The "temporary copy" idea is pretty central for visual models like Midjourney or DALL-E, whose training sets are full of copyrighted works lol. There is a legal basis for temporary copies too:
The "Ephemeral Copy" Exception (17 U.S.C. § 112 & § 117)
U.S. copyright law recognizes temporary, incidental, and transitory copies as necessary for technological processes. Section 117 allows temporary copies for software operation. Section 112 permits temporary copies for broadcasting and streaming. -
Okay, I can work with this. Hey Altman you can train on anything that's public domain, now go take those fuck ton of billions and fight the copyright laws to make public domain make sense again.
-
counterpoint: what if we just make an exception for tech companies and double fuck consumers?
Counter counterpoint: I don't know, I think making an exception for tech companies probably gives a minor advantage to consumers at least.
You can still go to copilot and ask it for some pretty fucking off the wall python and bash, it'll save you a good 20 minutes of writing something and it'll already be documented and generally best practice.
Sure the tech companies are the one walking away with billions of dollars and it presumably hurts the content creators and copyright holders.
The problem is, feeding AI is not significantly different than feeding Google back in the day. You remember back when you could see cached versions of web pages. And hell their book scanning initiative to this day is super fucking useful.
If you look at how we teach and train artists. And then how those artists do their work. All digital art and most painting these days has reference art all over the place. AI is taking random noise and slowly making things look more like the reference art that's not wholly different than what people are doing.
We're training AI on every book that people can get their hands on, But that's how we train people too.
I say that training an AI is not that different than training people, and the entire content of all the copyright they look at in their lives doesn't get a chunk of the money when they write a book or paint something that looks like the style of Van Gogh. They're even allowed to generate content for private companies or for sale.
What is different, is that the AI is very good at this and has machine levels of retention and abilities. And companies are poised to get rich off of the computational work. So I'm actually perfectly down with AI's being trained on copyrighted materials as long as they can't recite it directly and in whole, But I feel the models that are created using these techniques should also be in the public domain.
-
Copyright law doesn't cover recipes.
-
Actually I would just make the guard rails such that if the input can’t be copyrighted then the ai output can’t be copyrighted either. Making anything it touches public domain would reel in the corporations enthusiasm for its replacing humans.
I think they would still try to go for it but yeah that option sounds good to me tbh
-
Stealing means the initial item is no longer there
If someone is profiting off someone elses work, i would argue its stealing
-
Counter counterpoint: I don't know, I think making an exception for tech companies probably gives a minor advantage to consumers at least.
You can still go to copilot and ask it for some pretty fucking off the wall python and bash, it'll save you a good 20 minutes of writing something and it'll already be documented and generally best practice.
Sure the tech companies are the one walking away with billions of dollars and it presumably hurts the content creators and copyright holders.
The problem is, feeding AI is not significantly different than feeding Google back in the day. You remember back when you could see cached versions of web pages. And hell their book scanning initiative to this day is super fucking useful.
If you look at how we teach and train artists. And then how those artists do their work. All digital art and most painting these days has reference art all over the place. AI is taking random noise and slowly making things look more like the reference art that's not wholly different than what people are doing.
We're training AI on every book that people can get their hands on, But that's how we train people too.
I say that training an AI is not that different than training people, and the entire content of all the copyright they look at in their lives doesn't get a chunk of the money when they write a book or paint something that looks like the style of Van Gogh. They're even allowed to generate content for private companies or for sale.
What is different, is that the AI is very good at this and has machine levels of retention and abilities. And companies are poised to get rich off of the computational work. So I'm actually perfectly down with AI's being trained on copyrighted materials as long as they can't recite it directly and in whole, But I feel the models that are created using these techniques should also be in the public domain.
giving an exception to tech companies gives an advantage to consumers
No. shut the fuck up. these companies are anti human and only exist to threaten labor and run out the clock on climate change so we all die without a revolution and the billionaires flee to the bunkers they're convinced will save them (they won't, closed systems are doomed)
good for writing code
so, I have tried to use it for that. nothing I have ever asked it for was remotely fit for purpose, often referring to things like libraries that straight up do not exist.
AI
HOLY SHIT WE HAVE AI NOW!? WHEN DID THIS HAPPEN!? can I talk to it? or do you just mean large language models?
there's some benefit in these things regurgitating art
tell me you don't understand a single thing about how these models work, and don't understand a single thing about the value meaning or utility of art, without saying "I don't understand a single thing about how these models work, and don't understand a single thing about the value meaning or utility of art.".
